Learning constraint sub-hierarchies. The Bidirectional Gradual Learning Algorithm∗

It is a common feature of many case marking languages that some but not all objects are case marked.1 However, it is usually not entirely random which objects are marked and which aren’t. Rather, case marking only applies to a morphologically or semantically well-defined class of NPs. Take Hebrew as an example. In this language, definite objects carry an accusative morpheme while indefinite objects are unmarked. (1) a. Ha-seret her?a?et-ha-milxama the-movie showed ACC-the-war b. Ha-seret her?a (*?et-)milxama the-movie showed (*ACC-)war (from Aissen 2000) Similar patterns are found in many languages. Bossong (1985) calls this phenomenon “Differential Object Marking ” (DOM). A common pattern is that all NPs from the top section of the definiteness hierarchy are case marked while those from the bottom section are not